Logbook ยท 01 ยท Vibe coding

What I learned building software together with AI

The experience was extremely productive and constructive. I didn't take a "copy and paste" approach. It felt more like collaborating with a multidisciplinary team, fast at writing code from very focused, concise specifications โ€” one of the things I had to manage was resource consumption, in terms of tokens, which "forced" me to look for ways of interacting that minimized it.

This experience, which is only the beginning, let me see first-hand the huge potential of generative AI systems.

Vibe coding doesn't simplify the work โ€” it redefines it.
What I learned
  • โ†’AI massively speeds up writing code, but it can't be delegated the job of defining requirements, the spec, or responsibility for the final product.
  • โ†’Code is never right on the first pass. If it isn't organized under precise rules, it becomes unmanageable โ€” the help AI agents can give shrinks, to the point of vanishing. Using them without awareness is counterproductive.
  • โ†’Vibe coding doesn't simplify the work, it redefines it: you have to be aware of what you want to do, how you want to do it, what you need to check, and how you'll use it.
#vibecoding #buildinpublic #generativeAI
The project born from this experience, Registro, is now available on the Microsoft Store.
Next: Why I applied Agile to vibe coding โ†’